RADIA

Medical imaging AI demands both clinical accuracy and real-time performance — single-model approaches lack the depth of multi-specialist review.

• Single-model limitation — one LLM cannot match the coverage of multiple specialists reviewing the same image
• Latency vs depth — detailed analysis takes time, but clinicians need fast initial reads
• Modality diversity — X-ray, CT, MRI, ultrasound, mammography, and pathology each need specialized prompts

Dual-model architecture with parallel specialist consultations:

• Gemini 2.5 Flash — primary analysis engine for Scan360 12-point structured reports
• Groq (Llama 4) — parallel specialist in Asteroid Mode for independent second-opinion analysis
• AI image classification — pre-upload classify endpoint auto-detects study type and modality
• Hidden botany mode — auto-activates when plant images are detected, running a 12-point botanical analysis

Complete imaging AI pipeline:

• Scan360 — 12-point structured analysis: findings, impression, differential diagnosis, measurements, comparison, recommendations, clinical correlation, quality, limitations, follow-up, prognosis, teaching points
• Asteroid Mode — Gemini + Groq parallel execution with merged specialist-level output
• Auto-classify — AI endpoint classifies uploaded images by modality (XR, CT, MR, US, MG, pathology) before analysis begins
• Botany easter egg — hidden mode with species identification, phenological state, health assessment, disease/pest detection, nutritional analysis, edibility, and prognosis